This is for images in the frames folder only.
NOTE: Some species are heavily biased in the original dataset. They are not normalized in this dataset.
- Generate dataset JSON
python convert_to_json.py --reduce 100 --eval_p 15 --test_p 5 --min_ann 4 --frame_meta_csv_path "frame_metadata.csv" --middle
| Option | Description | Example | Default |
|---|---|---|---|
| --reduce | Percentage of dataset to use. Mainly to create smaller dataset for faster testing | --reduce 100 | 100 |
| --eval_p | Percentage to use for evaluation | --eval_p 15 | 15 |
| --test_p | Percentage to use for testing | --test_p 5 | 5 |
| --min_ann | Minimum annotations an image must have before being added to the new dataset | --min_ann 4 | 4 |
| --frame_meta_csv_path | Path to frame metadata csv | --frame_meta_csv_path "frame_metadata.csv" | "frame_metadata.csv" |
| --single_class | Group all species as 1 class. Useful when counting only. | --single_class | False |
| --middle | Convert to middle format. Setting false will leave the json in Detectron2 format | --middle | True |
- Download images in JSON. tqdm is required. Run "pip install tqdm". Will skip already downloaded files.
python download_dataset_from_json.py --json_path "output/middle_1324_C127_test_set.json" --base_url "https://data.pawsey.org.au/download/FDFML/frames/" --download_path "frames"
- (Optional) Copy test images out of the dataset.
python copy_json_images_to_folder.py --json_path "output/middle_1324_C127_test_set.json" --dataset_root "frames" --destination_root "curated_dataset" --max_files 60
- (Optional) Preview the dataset and annotations.
python preview_annotations.py --json_path "output/middle_1324_C127_test_set.json" --dataset_root "curated_dataset"